CrackML by @ml.with.umang
Interview questions / GenAI & LLM
GenAI & LLM interview question

What Is an LLM?

Explain what a large language model is to an ML engineer, including training and inference at a high level.

easyconceptEvidence 34/1001 source reportPinterest

The 60-second answer

An LLM is typically a large Transformer trained to model token sequences, most commonly by predicting the next token from previous context. Pretraining learns broad statistical representations from large corpora; instruction tuning/preferences can adapt behavior for conversational or task use.

Build the answer in this order

1
Define the mechanism

An LLM is typically a large Transformer trained to model token sequences, most commonly by predicting the next token from previous context.

2
Explain the architecture

Pretraining learns broad statistical representations from large corpora; instruction tuning/preferences can adapt behavior for conversational or task use.

3
Compare trade-offs

At inference the model repeatedly produces a probability distribution over the next token and decodes according to a sampling/selection strategy.

4
Close with serving + evaluation

Separate model knowledge from external grounding: tools/RAG can add current/private evidence without changing model weights.

A useful interview mental model

This is the shape of a strong answer—not a script to memorize.

01Definition
02Mechanism
03Trade-offs
04Failure modes
05When to use

Senior-level signal

  • Senior answers distinguish pretraining, supervised fine-tuning, preference optimization, and inference-time prompting/RAG.
  • Include context-window, latency, cost, hallucination, and evaluation constraints rather than defining an LLM only by parameter count.

What the interviewer is really testing

Understanding beyond prompting: architecture, retrieval, evaluation, inference, safety, latency, cost, and failure recovery.

Likely follow-up questions

What assumption makes this approach work?
When would you choose the strongest alternative instead?
What production or data failure mode changes your answer?

Common weak-answer patterns

  • Reciting a definition without mechanism or assumptions.
  • Claiming one technique is always better without a data regime.
  • Stopping before failure modes, validation, or deployment implications.